基于反应谱嵌入的高速铁路轨道-简支梁桥系统地震响应预测模型OA
Response spectrum-embedded seismic response prediction model for high-speed railway track-simply supported bridge system
震致轨道变形是影响震后高速铁路桥梁安全运维的最重要因素之一.为实现高速铁路轨道-简支梁桥系统震致位移响应快速预测,提出了一种基于时间卷积网络(temporal convolutional networks,TCN)与双向长短期记忆网络(bidirectional long short-term memory,Bi-LSTM)结合的深度学习模型,并将所提模型引入地震加速度反应谱与自注意力机制进行优化,以更精准地捕捉地震波特征与演化规律,显著提升地震响应预测的准确性与稳定性.针对优化前后模型在Ⅰ0场地下的预测数据开展了详细对比分析,评估了优化模型在钢轨、滑动层和CA(cement asphalt)砂浆层峰值位移预测精度上的提升效果.选取不同场地类别及地震动峰值加速度(peak ground acceleration,PGA)工况,系统探讨了模型在复杂环境下的预测性能变化规律.利用优化模型对不同场地类别、不同PGA下4 000条地震动对应的钢轨轨向峰值位移进行了预测,进一步探讨了钢轨轨向峰值位移与PGA和场地类别的关系.结果表明,优化模型在Ⅰ0类场地类别下对关键构件位移响应的预测精度较未改进模型有显著提升,其中均方根误差(frmse)最大降低超35%.优化模型在5种场地类别下的整体预测精度较高,决定系数(R2)稳定在0.92～0.99之间,表现出良好的鲁棒性.钢轨峰值位移最大值随着PGA的增大以及场地土体柔软程度的提升而增大.研究成果为高速铁路轨道-简支梁桥系统在地震作用下的位移响应预测提供了一种实用、可靠、有效的深度学习模型.
Seismic-induced track deformation is one of the most critical factors that affect the post-earthquake safety and maintenance of high-speed railway bridges.To enable rapid prediction of seismic-induced displacement responses in high-speed railway track-simply supported bridge system,this study proposed a deep learning framework that combined temporal convolutional networks(TCN)and bidirectional long short-term memory(Bi-LSTM)networks.The proposed model was further optimized by embedding the seismic acceleration response spectrum and incorporating a self-attention mechanism to more accurately capture seismic wave features and temporal evolution,thereby significantly enhancing prediction accuracy and robustness.A detailed comparative analysis was conducted between the optimized and baseline models under ClassⅠ0 site conditions to assess the improvement in peak displacement prediction accuracy for the rail,sliding layer,and cement asphalt(CA)mortar layer.Different site classes and peak ground acceleration(PGA)levels were considered to systematically investigate the variation of the model's predictive performance under complex conditions.Using the optimized model,4 000 seismic records across different site classes and PGA levels were analyzed to predict rail alignment peak displacements.It further revealed the relationships between rail displacement amplitude,PGA,and site classification.The results show that the optimized model significantly can improve prediction accuracy for key component displacement responses under Class Ⅰ0 site conditions,with the maximum reduction in root mean square error(frmse)exceeding 35%.The optimized model can achieve high overall prediction accuracy across five site classes,with the coefficient of determination(R2)consistently ranging from 0.92 to 0.99,demonstrating strong robustness.The maximum alignment rail displacement can increase with the increase in PGA and the softening of site soils.The proposed research can offer a practical,robust,and efficient deep learning model for predicting seismic-induced displacement responses of high-speed railway track-simply supported bridge system.
周旺保;刘昊;彭康;彭梦颖;徐杰玉;秦浩然
中南大学 高速铁路建造技术国家工程研究中心,湖南 长沙 410075||中国中铁股份有限公司,北京 100039||中南大学 土木工程学院,湖南 长沙 410075中南大学 高速铁路建造技术国家工程研究中心,湖南 长沙 410075||中南大学 土木工程学院,湖南 长沙 410075中南大学 高速铁路建造技术国家工程研究中心,湖南 长沙 410075||中南大学 土木工程学院,湖南 长沙 410075中南大学 高速铁路建造技术国家工程研究中心,湖南 长沙 410075||中南大学 土木工程学院,湖南 长沙 410075中南大学 高速铁路建造技术国家工程研究中心,湖南 长沙 410075||中南大学 土木工程学院,湖南 长沙 410075中南大学 高速铁路建造技术国家工程研究中心,湖南 长沙 410075||中南大学 土木工程学院,湖南 长沙 410075
交通工程
深度学习高速铁路轨道-简支梁桥系统地震响应峰值位移
deep learninghigh-speed railwaytrack-simply supported bridge systemseismic responsepeak displacement
《铁道科学与工程学报》 2026 (7)
3047-3058,12
中国中铁股份有限公司科技研究开发计划项目(2022-重大-17)国家自然科学基金资助项目(52178180)湖南省科技人才托举工程(2022TJ-Y10)中南大学前沿交叉研究项目(2023QYJC006)澳门特别行政区科学技术发展基金资助项目(SKL-IOTSC(UM)-2024-2026)
评论